Introduction 4

 Scope and audience 4

 What is agentic AI? 5

 How is it different from generative AI? 5

Broader agentic AI security considerations 6

 Inherited risks of LLMs 6

 Increased attack surface 6

 Increased complexity 6

 Evolving security as technology matures 7

 AI security as part of cyber security 7

Agentic AI security risks 7

 Privilege risks 7

 Design and configuration risks 9

 Behaviour risks 9

 Structural risks 11

 Accountability risks 12

Best practices for securing agentic AI systems 14

 Designing secure agents 14

 Developing secure agents 16

 Deploying agents securely 18

 Operating agents securely 21

Defend against future risks 24

 Expand threat intelligence through collaboration 24

 Develop robust, agent-specific evaluations 24

 Leverage system-theoretic approaches to analyse security 25

Conclusion 25

Further information  26

Appendix A 28

 Cyber security prerequisites before implementation of AI agents 28